Lab Sessions- Mean: the average score, add all the scores together and divide by # of scores- Median: the score in the middle- Mode: the most frequent score- Standard deviation: the square root of the variance, deviation from the mean- Variance (s^2): the sum of squares divided by sample size, the spread of the numbersVariance- The components of variance in your data: - Total variance = systematic variance + error variance- Total variance = observed differences- Systematic variance: effect of IV- Error variance: variability in scores caused by extraneous variables or participant variability (effect of all other things)- Why is error variance a problem?- Could show effect of IV, when in fact there wasn’t one, or could show a greater effect than there actually is- Where does error variance come from?- Extraneous scores or participant variability- Regularize experiment to remove error variance- Reduce sources of error variance- Four ways to deal with error variance:- Control it- Treat all participants the same- Match participants on variables that have an effect on DV- Randomize it across conditions- Make it so participants have an equal chance of being in any group- Increase effectiveness of your IV- Use levels of IV that are very different- Account for it with statistics- Estimate the probability that error variance caused the observed effectDesigns- Randomized 2-Group Designs- Advantages: Simple, few participants required, easy to interpret, no pre-testrequired to ensure equality of groups- Disadvantages: doesn’t yield a large amount of info, insensitive to effects when participants differ greatly in performance- Randomized Multi-Group Designs- Properties: comparing two or more treatments to one or more control groups- Advantages: permit comparing two or more treatments to one or more control groups- Disadvantages: multiple control groups necessary to rule out alternative explanations- Matched Pair- Properties: Measure sample and find pairs of people who match on characteristics that might influence performance, for each pair randomly assign which person goes into which group- Matched Multi-Group Designs- Properties: 3 or more groups, find similar participants than randomly assign these participants to groups- Advantages: help to control for error variance by matching groups on characteristics that influence performance- Disadvantages: difficult to implement, require use of different statistics- Within-Participant Designs- Properties: same subjects used in all conditions- Advantages: reduces error variance due to individual differences among subjects across treatment groups, requires fewer participants- Disadvantages: more demand put on participants, carry over effects- Factorial Designs- Properties: more than one IV, use either between or within participant IVs-Advantages: more efficient, greater external validity, can test main effects and interactions-Disadvantages: more complex, more participants, difficult to interpret*Repeated-measures is another way to say within-subjectsMatched Design- Advantages:- Controls participant- related variability- Matches groups on characteristics that influence performance- Systematic variance is easier to observe in this design- Hurts ability to detect an effect- Need to measure all participants before study- Many groups = difficult to find people who matchCarryover Effects- Exposure to a previous treatment affects performance in a subsequent treatment- Ways to deal with them:- Minimize their effect: introduce breaks into the study to allow effect of previous treatment to wear off, practice trials- Counterbalance: varying the order of treatments- Advantage: all possible combinations of treatments are represented- Disadvantage: minimum number of subjects can get large fast- Include treatment order as an IV: will let you know if significant carryover effects are present in your study- Carryover effects can sometimes not be dealt with:- When treatments have irreversible effects- Different carryover effects (Example: Treatment A has a different effect on B than B does on A)Methods of Counterbalancing- Partial Counterbalancing: includes some of the possible treatment orders but not all - Reverse counterbalancing: use with small number of conditions- Random Order counterbalancing: assign random order of treatment to each participant- Latin square design- Controls for ordinal position of each treatment- Each condition follows each other condition onceFactorial Design- Properties- Has more than 1 IV- Can use within or between participant IVs- Main effect: the separate effect of each independent variable- Main effect = the number of factors- Interaction: the effect of one IV changes based on the level of another IV- Signature of an interaction graphically:- Interaction if the two lines are parallel or perpendicular (cross in the middle)Nomenclature of Factorial Designs- Factor: each IV- Levels: each factor has a number of levels (treatments)- Condition: a specific combination of levels- In a given design:- Number of potential main effects = # of IVs- Number of interactions = number of possible combinations of IVs-Example: 2 x 3 x 2 = 3 factors, 7 levels, 12 conditions, 3 main effects, 4 interactionsSpecific Factorial Design- Participants receive all the treatments of one IV and only some of the other: MIXEDdesign- Main limitation of a factorial design when the number of IVs is 3 or more: difficult to interpretCovariate- Combining correlation and experimental methods- The idea of holding something constant to reduce error variance- Measuring something to get rid of it- Example: If there is a systematic relationship between IQ and our measure, we should be able to predict what everyone’s score would have been if they were all thesame IQ.Quasi-independent Variable- Including measured or subject variables in your experiment- IV is not manipulated- Advantages of including one in your study: - We can look at the generalizability of a result- Like a covariate, it can suck up some of the variance in the data and allow usto see the effect of our IV by grouping together participants who behaved similarly- Biggest problem associated with a quasi-IV:- Causation cannot be inferred (whenever a subject variable is in your model you cannot infer causation)- Threat to internal validityBook ObjectivesChapter 8- Inferential Statistics- Tell us if the difference between groups is larger than expected due to chance or error variance- Used to test a research
View Full Document